Improved Cuckoo Search Algorithm for Hyper-parameter Optimization in Deep Learning and Its Application in Activity Recognition

Abstract Current research shows that deep learning algorithm is an effective method for activity recognition. However, the deep learning algorithms often contain some hyper-parameters which may be continuous, integer, or mixed, and are often given based on experience but largely affect the effectiveness of activity recognition. In this paper, Cuckoo Search (CS) algorithm is proposed to optimize hyper-parameters in deep learning models. In order to adapt to different hyper-parameter optimization problems, an improved CS algorithm is proposed. Then, the hyper-parameters in CNN and LSTM are optimized based on the improved CS algorithm on the smart home activity recognition dataset. Recognition results show that hyper-parameter optimization based on improved CS algorithm can improve the performance of the deep learning model.